Обсуждение: pg_repack issues

-----BEGIN PGP SIGNED MESSAGE-----
Hash: RIPEMD160


Mark Steben wrote:
> ERROR: query failed: ERROR:  value "3048471620" is out of
> range for type integer
> DETAIL: query was: SELECT pg_try_advisory_lock($1, $2)
>
> This happens because the oid values of the tables is this database are
> all over 2 GB.  Can I get some insight to make pg_repack work in this
> situation?

That was an unfortunate choice they made in having their second (int)
argument be an OID. A quick fix would be to have it use the single (bigint)
form of pg_try_advisory_lock. Something like this (untested):

diff --git a/bin/pg_repack.c b/bin/pg_repack.c
index af510ca..d638164 100644
- --- a/bin/pg_repack.c
+++ b/bin/pg_repack.c
@@ -1520,13 +1520,12 @@ static bool advisory_lock(PGconn *conn, const char *relid)
 {
    PGresult       *res = NULL;
    bool            ret = false;
- -   const char     *params[2];
+ const char     *params[1];

- -   params[0] = REPACK_LOCK_PREFIX_STR;
- -   params[1] = relid;
+ params[0] = relid;

- -   res = pgut_execute(conn, "SELECT pg_try_advisory_lock($1, $2)",
- -                      2, params);
+ res = pgut_execute(conn, "SELECT pg_try_advisory_lock($1)",
+                    1, params);

    if (PQresultStatus(res) != PGRES_TUPLES_OK) {
        elog(ERROR, "%s",  PQerrorMessage(connection));




You also may want to raise this as an official issue here:

https://github.com/reorg/pg_repack/issues

- --
Greg Sabino Mullane greg@turnstep.com
End Point Corporation http://www.endpoint.com/
PGP Key: 0x14964AC8 201411112120
http://biglumber.com/x/web?pk=2529DF6AB8F79407E94445B4BC9B906714964AC8
-----BEGIN PGP SIGNATURE-----

iEYEAREDAAYFAlRixSsACgkQvJuQZxSWSsjuFwCfRvWEMeP3Rnb7F0HuRvIMNKwi
wNwAoPq3Sg6Q64C2tK/FdLdTHBo3zqlm
=DiA9
-----END PGP SIGNATURE-----

Just happened to see this thread -- the github issues page is the
proper place to send bug reports about pg_repack, and OP has opened
https://github.com/reorg/pg_repack/issues/30 already.

On Tue, Nov 11, 2014 at 9:26 PM, Greg Sabino Mullane <greg@turnstep.com> wrote:

> That was an unfortunate choice they made in having their second (int)
> argument be an OID. A quick fix would be to have it use the single (bigint)
> form of pg_try_advisory_lock. Something like this (untested):

[patch to use single-argument form of pg_try_advisory_lock]

Yeah, that would work. The reason we're using the two-argument form of
pg_try_advisory_lock() at all is an attempt to play nice with any
other applications which might also be using advisory locks; maybe
that's being excessively cautious. I don't have a strong preference,
but I believe this should also work:

SELECT pg_try_advisory_lock($1, -2147483648 + $2)

i.e. cram the unsigned 4-byte OID into a signed 4-byte integer.

Josh